TECHDESK DAILY — Pre-Market Edition · Wednesday, September 23, 2026

Pre-market: prices are the Tuesday September 22 close plus any pre-market quotes. · 109 stories reviewed

Top Stories

Boeing, GoGo test satellite-linked military drone in challenge to Starlink CONSENSUS

Boeing and GoGo are trialing a satellite‑linked military drone that could compete with SpaceX’s Starlink for defense communications, according to Reuters 【https://news.google.com/rss/articles/CBMixAFBVV95cUxOWWlfdHJPLVFJRU5BSkxDU3JfaG5lVkMyaFJLd1o1dEVub0RJRTlqMzRMd09nVEpxM3VtWFc4UTlzZ0xHY2ZSV2pibXlJSGxNR3hKRnI5V2VYcUwtUld2NVp5b250b1JEOEFPeWVycWlxYjZqQkNBeVVXSG5nT1F3cmc1OU41TDNtUzlNTEhFbzJ3eTVQckZNYjNlRm1EUXJLNkktazQ4RkJpSjgtbDFoUmI5WUZUY0pRaElxM20tYmlJVS1j?oc=5】. The test ties into a broader AI‑defense push, linking satellite‑linked drones with AI‑enhanced fighter jets (see s45 connection).
Desk take: Early insight into a potential shift in military communications that could affect defense contractors and satellite operators.

Nvidia's 800V power shift hands Taiwan's capacitor makers a new AI segment UNDER-COVERED

Nvidia’s rollout of an 800 V high‑voltage platform for AI data centers is spurring demand for advanced capacitors, opening a niche market for Taiwanese makers 【https://www.digitimes.com/news/a20260922PD237/demand-mlcc-holy-stone-prosperity-dielectrics-gpu.html】. Nvidia also launched the DSX Ready qualification program for power and cooling hardware 【https://www.servethehome.com/nvidia-announces-dsx-ready-qualification-program-for-data-center-power-and-cooling-hardware/】, and LG Energy’s BESS entered Nvidia’s AI Factory ecosystem 【https://www.digitimes.com/news/a20260923VL213/lg-nvidia-data-market-certification.html】. This aligns with the edition note linking s11, s27 and s42 on AI compute supply‑chain impacts.
Desk take: Watch Taiwanese capacitor firms for upside as Nvidia’s high‑voltage AI push gains traction.

Memory chips are now more expensive than compute chips on a per-area basis ▲ LOW DISSEMINATION

DRAM wafers now command higher per‑area pricing than TSMC’s leading‑edge N3 compute wafers, reflecting AI‑driven memory demand 【https://www.tomshardware.com/pc-components/dram/dram-is-now-more-expensive-than-compute-chips-on-per-area-basis-ai-demand-drives-memory-die-value-past-leading-edge-silicon】.
Desk take: Memory suppliers may see margin expansion, while chip fabs could face pricing pressure.

China's BeiDou eyes US$150 billion industry scale with AI, services at the core ▲ LOW DISSEMINATION

BeiDou’s satellite navigation ecosystem is expanding into chips, modules and end‑devices, with AI increasingly embedded in its services 【https://www.digitimes.com/news/a20260923PD228/beidou-chips-data-2025-hardware.html】.
Desk take: Chinese navigation assets could become a strategic AI platform, worth monitoring for supply‑chain exposure.

Dassault Aviation tests AI algorithms on Rafale fighter jet ▲ LOW DISSEMINATION

Dassault is evaluating AI software on its Rafale fighter, aiming to enhance performance and decision‑making 【https://news.google.com/rss/articles/CBMiuwFBVV95cUxNZUxwak1sd0hYUGdPdW16RUxHRkR0eXg1MmVmUzBLSkJVRWptcUUxUHRmX0Z6d3BFUmNTM3c0TmNKNXp2eWVYNmlvZlNsVkpQektGcG15cExLOHhibmNkVldkODhNWmdOQ0lsUHdNRXM0WWsyTUl0dFVqbUx4YXhKNWhteDNTVU9lUEMzeUR3ZDBCVWR6cG1RYkRZV2FXRkVBN2ZzX1d5MGo1aVdrcnc0VTFCVmdwQlZBcEEw?oc=5】. This links with s3’s AI‑defense narrative.
Desk take: AI integration in combat aircraft could reshape defense tech spend and vendor landscapes.

‘We’ll be the first to mine Bitcoin in space’ says Nvidia-backed startup ▲ LOW DISSEMINATION

StarCloud, backed by Nvidia, plans to launch a Bitcoin mining operation in orbit later this year 【https://www.tomshardware.com/tech-industry/cryptomining/well-be-the-first-to-mine-bitcoin-in-space-says-nvidia-backed-startup-starcloud-will-fire-up-its-asics-after-its-second-spacecraft-reaches-orbit-later-this-year】.
Desk take: Space‑based crypto mining introduces a novel revenue stream and potential regulatory scrutiny.

Kairos Power selects Samsung C&T to help build 50MW SMR for Google in Tennessee ▲ LOW DISSEMINATION

Kairos Power has partnered with Samsung C&T to construct a 50 MW small modular reactor for Google, slated for completion by 2030 【https://www.datacenterdynamics.com/en/news/kairos-power-selects-samsung-ct-to-help-build-50mw-smr-for-google-in-tennessee/】. This ties into the AI‑infrastructure theme linking s40 and s57.
Desk take: Google’s nuclear partnership signals a push for low‑carbon AI compute power, affecting energy and tech equities.

Google inks deal to support nuclear uprates at Georgia Power's Vogtle and Hatch plants ▲ LOW DISSEMINATION

Google will add 96 MW of capacity by supporting uprates at the Vogtle and Hatch nuclear facilities 【https://www.datacenterdynamics.com/en/news/google-inks-deal-to-support-nuclear-uprates-at-georgia-powers-vogtle-and-hatch-plants/】, reinforcing its strategy to secure clean energy for AI workloads.
Desk take: Expanding nuclear capacity underpins Google’s AI compute ambitions, highlighting energy‑tech convergence.

Under the Radar

• ▲ LOW DISSEMINATION — Microsoft to give DARPA hands‑on access to quantum system at new Maryland site: Microsoft will allow DARPA direct access to its quantum computing system at a newly announced Maryland facility 【https://news.google.com/rss/articles/CBMiuAFBVV95cUxNOTVRMXpJeWMzYnQzYnFEQ2JRdEI5X1RHaDVTRDNrYlhkbXRNNThYZDhiZWhpalA5dUEweGthNy1EdEVOTk45MXJxbG5mcXNSbjZHeHc1aUFmaDJVbFNFU290RlFwOGlXdkVvcENHVk0yNmUtRlFpS01DeWJuN1JPdU9fOXlyN2J6THJtRkxBWG5aSDRCcTR4d1o1ZXNfamh5VzlQM0k3ZHh4RDRHc05Cbk81bzNQX2xY?oc=5】. Desk take: quantum access could accelerate defense‑related tech pipelines.
• ▲ LOW DISSEMINATION — Topology‑Aware Workload Scheduling with NVIDIA Topograph: NVIDIA’s blog details a new topology‑aware scheduler that improves GPU placement in power‑limited AI factories 【https://developer.nvidia.com/blog/topology-aware-workload-scheduling-with-nvidia-topograph/】. Desk take: tighter scheduling can lift performance without extra hardware spend.
• ▲ LOW DISSEMINATION — Accelerating a ROS 2 Node with an AI Agent and NVIDIA Isaac ROS: NVIDIA explains how adding an AI agent and Isaac ROS can speed up compute‑intensive robotics workloads beyond raw CUDA kernels 【https://developer.nvidia.com/blog/accelerating-a-ros-2-node-with-an-ai-agent-and-nvidia-isaac-ros/】. Desk take: robotics teams can gain latency wins using existing GPU stacks.
• ▲ LOW DISSEMINATION — Enabling Private High‑Performance Production AI Inference with NVIDIA Confidential Computing: NVIDIA outlines confidential‑computing techniques that let LLM inference run on sensitive data while preserving performance 【https://developer.nvidia.com/blog/enabling-private-high-performance-production-ai-inference-with-nvidia-confidential-computing/】. Desk take: enterprises can adopt LLMs without exposing proprietary data.
• ▲ LOW DISSEMINATION — Smarter Buildings, Safer Occupants: Intelligence Meets Privacy: EE Times reports on building‑automation systems that balance energy efficiency with occupant health, safety and privacy 【https://www.eetimes.com/smarter-buildings-safer-occupants-intelligence-meets-privacy/】. Desk take: IoT‑enabled real‑estate assets may see higher ESG scores.
• ▲ LOW DISSEMINATION — What’s New for Game Developers: DLSS 5 with 3D‑Guided Neural Rendering, NVIDIA ACE Updates, and New RTX Kit Capabilities: NVIDIA introduces DLSS 5, adding 3D‑guided neural rendering and finer control for developers 【https://developer.nvidia.com/blog/whats-new-for-game-developers-dlss-5-with-3d-guided-neural-rendering-nvidia-ace-updates-and-new-rtx-kit-capabilities/】. Desk take: next‑gen visual fidelity could boost game‑engine licensing demand.
• ▲ LOW DISSEMINATION — 微软 Xbox 再裁 268 人，动视全面接手《光环》系列: TechNews Taiwan reports Xbox laid off 268 positions as it hands the Halo franchise to Activision 【https://ccc.technews.tw/2026/09/23/xbox-lays-off-268-more-people-hands-halo-series-to-activision/】. Desk take: cost cuts may free cash for Microsoft’s broader cloud and AI bets.
• ▲ LOW DISSEMINATION — BYD Semiconductor Strategy Advances With Localization: thebuzzevnews.com notes BYD Semiconductor is localising its supply chain and production 【https://news.google.com/rss/articles/CBMiggFBVV95cUxPNDM1OUZfVXRJWFAtZV84dGk5R2J0c3dTNlpSS0lhMzVVeHNSNGo4dXMxMVBSOC1vYjBoOFpaMF9JRjhiWXl3OEYtbjdqY2NJc2d6TE15STFMSUY3cmZlbkEyRjNJSVhzVnpKTDJjOVhqS2xyd0pMS21XQlI1cUpSVU1B?oc=5】. Desk take: localized chip production could reduce exposure to export controls.
• ▲ LOW DISSEMINATION — Travel Group TUI Tightens Earnings Guidance, Citing Strong Vacation Demand: WSJ reports TUI raised its earnings guidance on robust demand 【https://news.google.com/rss/articles/CBMivAFBVV95cUxOZW9hNmxmZVhDelBvOUpxeFNWVmlVUVBnWXpOSnJrYnoxd1F3OURXYXA5b0RKMHpxOTh0X2gzYV9hSTdtWU8xdlU3VUdwNTViZ3YzTkQ1M3FWUW5RT0xoLWpncXVyOFdMaGlXcU9UQ2Vla0NtM0FRa2NJMXZieGNhc0ZJSjBYVDBoLV84N2txWngwcjFkSzVKWmFtZ0x0Q0owY1JpSnBKcnVGTVMtRF9QeVhKZEl3c2g4Mi12bQ?oc=5】. Desk take: leisure‑travel exposure may add upside to consumer‑discretionary allocations.
• ▲ LOW DISSEMINATION — Lawsuit demands OpenAI pay for new school after ChatGPT used in shooting: Ars Technica details a British Columbia lawsuit seeking damages from OpenAI over a shooter’s ChatGPT logs 【https://arstechnica.com/tech-policy/2026/09/lawsuit-demands-openai-pay-for-new-school-after-chatgpt-used-in-shooting/】. Desk take: legal risk highlights the need for tighter content‑use safeguards.
• ▲ LOW DISSEMINATION — OpenAI wants to consult elite mathematicians about how to not fumble again: The Verge reports OpenAI is assembling a panel of top mathematicians after a reputational crisis over flawed math results 【https://www.theverge.com/ai-artificial-intelligence/999167/openai-elite-mathematicians-panel】. Desk take: expert advisory could restore credibility for OpenAI’s research pipeline.
• ▲ LOW DISSEMINATION — Rabbit’s new AI agent doesn’t need an R1 to run: The Verge notes Rabbit is launching a standalone AI agent that operates without its R1 hardware 【https://www.theverge.com/ai-artificial-intelligence/999094/rabbit-ai-agent-os3】. Desk take: hardware‑agnostic AI services may broaden Rabbit’s market reach.

Conversational Intelligence

• Adam Foroughi, CEO, Applovin (All-In, 2026-09-20): Applovin's share price rose from $80 to $150 in one week after the CEO began talking to investors in September 2023. “In that week, the stock went from 80 to 150.” [min 13]. Data unavailable — dissemination tag. Desk take: rapid short‑term price moves signal market sensitivity to executive communication.
• unattributed (Dwarkesh Podcast, 2026-09-17): A model deployment system updates and deploys a new model every five hours if it improves on the cursor benchmark. Paraphrased from discussion. [min 44]. Data unavailable — dissemination tag. Desk take: frequent model turnover could accelerate AI product iteration cycles.
• Stefano Ermon, CEO, Inception (No Priors, 2026-09-18): Inception is currently serving diffusion‑based LLMs in production with a custom serving engine. “We are serving these models in production today... we had to build our own serving engine.” [min 13]. Data unavailable — dissemination tag. Desk take: proprietary serving stack may create a competitive moat.
• Adam Foroughi, CEO, Applovin (All-In, 2026-09-20): Applovin's share price moved from $9 to $750 over two and a half years, taking market cap from $3.8 billion to $250 billion. “We ended up going from nine to $750 a share in a matter of two and a half years. ... we had a $3.8 billion market cap ... we got to $250 billion market cap.” [min 15]. Data unavailable — dissemination tag. Desk take: long‑term equity appreciation underscores strong growth trajectory.
• unattributed (Dwarkesh Podcast, 2026-09-17): A fine‑tuned Taki model, originally trained on data only up to 1930, outperformed Claude III on the SweepBench coding benchmark. Paraphrased from discussion. [min 62]. Data unavailable — dissemination tag. Desk take: legacy‑trained models can still achieve state‑of‑the‑art performance when fine‑tuned.
• Stefano Ermon, CEO, Inception (No Priors, 2026-09-18): Inception was founded two years ago and now has about 50 employees, most of whom focus on R&D. “We are about two years old, around 50 people, spending a lot of time still on R&D.” [min 13]. Data unavailable — dissemination tag. Desk take: small, R&D‑heavy team suggests early‑stage scaling potential.
• Adam Foroughi, CEO, AppLovin (All-In, 2026-09-20): AppLovin's EBITDA margin is 84%, the highest in the market. “our EBITDA margins, I think, are number one in the market. It's 84%.” [min 21]. Data unavailable — dissemination tag. Desk take: superior profitability enhances valuation resilience.
• unattributed (Dwarkesh Podcast, 2026-09-17): Cumulative improvement in model performance has been roughly 27×, corresponding to about threefold gains per year since 2019, extrapolating to a 3⁷ (~2,000×) total improvement. Paraphrased from discussion. [min 67]. Data unavailable — dissemination tag. Desk take: sustained exponential gains indicate continued upside for AI capabilities.
• Stefano Ermon, CEO, Inception (No Priors, 2026-09-18): The main bottleneck for Inception is compute resources rather than headcount. “But I think the bottleneck is more compute than people.” [min 34]. Data unavailable — dissemination tag. Desk take: compute constraints may limit near‑term scaling despite ample talent.
• Chamath Palihapitiya, host (All‑In, 2026-09-19): 45 days after the first 737 MAX crash, almost everyone knew the MCAS software was at fault but neither Boeing nor the FAA grounded the plane. “45 days after the first crash, almost everyone knew the MCAS software was at fault, and no one grounded the plane.” [min 8]. Data unavailable — dissemination tag. Desk take: regulatory lag poses risk to aerospace equities.
• unattributed (Dwarkesh Podcast, 2026-09-17): Frontier open‑source models have been scaling parameters at roughly 2× per year. Paraphrased from discussion. [min 70]. Data unavailable — dissemination tag. Desk take: rapid parameter growth may fuel open‑source competition.
• Stefano Ermon, CEO, Inception (No Priors, 2026-09-18): Inception cannot disclose model size or FLOPs, describing them as a trade secret. “We are not able to share much about the training, the size of the models or the flops or all of that. It's kind of like a trade secret.” [min 32]. Data unavailable — dissemination tag. Desk take: opacity could affect investor assessment of tech depth.

Tape Check

Snapshot as of 2026-09-23 08:40 UTC.

Today / Watch

AI is reshaping high‑value, under‑reported infrastructure—from defense and energy to space‑based finance—driving new supply‑chain and regulatory dynamics.

109 stories were gathered and ranked this morning; the full ranked listing appears in The Full Tape below.